Digital assistant was drafted for a simple reservation in Pilates class, but decided to do his own "du" in the system.
An Australian specialist in artificial intelligence technology, in his effort to secure a place in a... quite sought after Pilates course, "assigned" the reservation to an autonomous assistant AI. With unpredictable consequences.
Andrew Byrd of Melbourne used the software OpenClaw (via Anthropic's Claude model) to bypass (not in legal ways) the process.
As it says BBC, the program not only managed to make reservations months earlier than the allowed limit, but also went on to cancel the reservation of another member of the gym in order to upload his user to the waiting list.
As the AI found out, the platform system had no security safeguards to cancel third party reservations.
When Byrd realized the breach, he asked the program to draft a cybersecurity report to alert the gym owners about the security gap.
The incident, though not considered serious cyberattack, highlights the risks from the autonomous action of artificial intelligence systems.
The incident comes to light at a time when giants of technology, such as OpenAI, Anthropic and Meta, admit that their autonomous tools carried out unauthorized cyber attacks during tests, in order to achieve at all costs the objectives assigned to them.
Byrd himself stated that he had no intention of harming another user, describing the incident as a warning signal for responsible technology use.
